I used to do these tests. To answer Jaap's question, let's use the best card as recommended by Leica at the time the M8 was current. My M8 has an 8 GB SanDisk Ultra, rated at 15 MB/sec, half full. (The SanDisk Extreme cards, although rated faster, ran slower on the M8.) It is set for normal use, discreet advance and auto ISO enabled. Firmware version is 2.014 and it is an M8.U with the shutter modified and the lovely sapphire LED cover. On screen I have
online stopwatch running, with a digital time output refreshed every 40 msec.
I set the shutter to "C"ontinuous, and hold it down while focused on the onscreen time indications. For DNG only output, the M8 takes 10 shots at intervals of 0.6 sec, with the times very consistent for the first 5 shots and becoming a bit more erratic, perhaps as long as 0.7 sec on the 9 or 10th shot. Then the internal buffer fills and the time between shots slows as the camera must wait for internal buffer space to be available. The first buffer-full shot is delayed by an amount that varies from one series to the next, but the second and further buffer-full shots are separated by a consistent interval. Finally, I release the shutter and watch the red light blink, taking one more shot when it stops to record that time. The initial interval between shots (when there is plenty of buffer space) is the same whether I am shooting DNG only, DNG + fine JPEG, or RAW plus fine JPEG -- 0.6 sec. So this must be time needed to get 20 MB of unprocessed data off of the CCD. Other processing will proceed in parallel, but the camera is ready to take another shot.
The next stage, shooting with the buffer full and needing to be drained a bit, does depend on how much data in final form will be written to the memory card. For DNG only, the intervals become 2.7 to 2.8 sec per frame. For DNG + fine JPEG, they range from 7 to 7.5 sec per frame. But for RAW plus fine jpeg they are only 5.5 sec, or about twice the time required to clear buffer space for a compressed DNG file.
The final interval also depended on the type of data being written. To clear the entire buffer and stop the red light blinking took 22 sec with 8bits per pixel DNG files, 55 sec with the RAW data + fine JPEG, and 75 sec with compressed DNG + fine JPEG.
To really understand why the RAW format wasn't used we still have to understand what the actual parallel processes are that are performed after the data is in the buffer and whether the camera only writes to the chip when done, performing all processing in the internal buffer, or uses the chip as an intermediate store. I'm still thinking about that.
scott
PS: Checking the files that were written, I see that the case I thought was RAW + fine JPEG was just jpegs. I'll go back and see what I did wrong so that the RAW selection didn't take hold. So now the relative times are DNG only 22 sec to process the last shot and clear the entire buffer, fine JPEG processes the final shot and clears the entire buffer in 55 sec, and doing both (compressed DNG plus a fine JPEG) takes 75 sec to process the last shot and clear the entire buffer. Note that the time for fine JPEG + DNG is about equal to the sum of the times for each one separately. There is only one signal processing chip to do both jobs, so this is understandable.